Smart Grid and Renewable Energy, 2012, 3, 89-95
http://dx.doi.org/10.4236/sgre.2012.32013 Published Online May 2012 (http://www.SciRP.org/journal/sgre) 1
Generation Reliability Evaluation in Deregulated Power
Systems Using Game Theory and Neural Networks
Hossein Haroonabadi1, Hassan Barati2
1Electrical Department, Islamic Azad University (Islamshahr Branch), Tehran, Iran; 2Electric al Department, Isla mic Azad University
(Dezful Branch), Dezful, Iran.
Email: haroonabadi@iiau.ac.ir, barati216@gmail.com
Received March 30th, 2012; revised April 27th, 2012; accepted May 4th, 2012
ABSTRACT
Deregulation policy has caused some changes in the concepts of power systems reliability assessment and enhancement.
In the present research, generation reliability is considered, and a method for its assessment is proposed using Game
Theory (GT) and Neural Networks (NN). Also, due to the stochastic behavior of power markets and generators’ forced
outages, Monte Carlo Simulation (MCS) is used for reliability evaluation. Generation reliability focuses merely on the
interaction between generation complex and load. Therefore, in the research, based on the behavior of players in the
market and using GT, two outco mes are con sider ed : co op eration and non-co op eratio n. Th e proposed meth od is assessed
on IEEE-Reliability Test System with satisfactory results. Loss of Load Expectation (LOLE) is used as the reliability
index and the resu lts show generation reliability in cooperation market is better than non-coop eration outcome.
Keywords: Power Market; Generation Reliability; Game Theory (GT); Neural Networks (NN); Monte Carlo
Simulation (MCS)
1. Introduction
Power systems have evolved over decades. Their primary
emphasis is on providing a reliable and economic supply
of electrical energy to their customers [1]. A real power
system is complex, highly integrated and almost very
large. It is divided into appropriate subsystems or func-
tional zones that can be analyzed separately [1]. This
paper deals with generation reliability assessment (Hier-
archical Level I-HLI) in power pool market, and the trans-
mission and distribution systems are considered reliable
and adequate as shown in Figure 1.
Most of the methods used for generation reliability
evaluation are based on the loss of load or energy ap-
proach. One of the suitable indices that describes genera-
tion reliability level is “Loss of Load Expectation” (LOLE),
that is the time in which load is more than the available
generation capacity.
Reliabl e Transmis-
si on & Di stribution
Systems
Ge n. 1
Ge n. 2
Ge n. n
Load
Figure 1. Power pool market schematic for HLI reliability
assessment.
Generally, the reliability indices of a system can be
evaluated using one of the following two basic approaches
[1]:
Analytical techniques,
Stochastic simulation.
Simulation techniques estimate the reliability indices
by simulating the actual process and random behavior of
the system. Since power markets and generators’ forced
outages have stochastic behavior, Monte Carlo Simula-
tion (MCS), as one of the most powerful methods for
statistical analysis of stochastic problems, is used for
reliability assessment in this research.
Since the beginning of the 21st century, many countries
have been trying to deregulate their power systems and
create power markets [2,3]. In the power markets, the
main function of players is their own profit maximization,
which severely depends on the type of the market. As a
result, generation reliability assessment depends on the
market’s type and characteristics.
Reliability problems have been evaluated in power
markets during the last decade [4,5]. This paper deals
with generation reliability in power pool markets using
game theory (GT) and Neural Networks (NN). GT is the
mathematical study of interaction among independent,
self-interested agents [6]. That is, where the actions of
one agent affect the payoff (utility or profit) of another
agent in a way that affects the choice of best action by
Copyright © 2012 SciRes. SGRE
Generation Reliability Evaluation in Deregulated Power Systems Using Game Theory and Neural Networks
90
the affected agent. In Section 2, fundamentals of GT and
its application to economics are discussed. Section 3,
deals with the algorithm for generation reliability assess-
ment in power markets using GT, and finally, case study
results are presented and discussed in Section 4.
2. Oligopoly Power Markets and GT
Concepts
Generally, economists divide the markets into four groups
[7]:
Perfect competition market,
Monopoly market,
Monopolistic competition market,
Oligopoly market.
Oligopoly market is a market in which the number of
buyers is small, while the number of sellers could be large.
This market differs from perfect competition market,
because each firm is large enough to have a significant
effect on the market. Also it differs from monopoly mar-
ket, because there is more than one firm in the market. It
differs from monopolistic competition market too, be-
cause its products are similar together.
Unlike a pure monopoly or perfect competitive firm,
most firms must consider the likely responses of com-
petitors when they make strategic decisions about price
advertising expenditure, investment in new capital and
other variables. The main question of each player is: If I
believe that my competitors are rational and act to maxi-
mize their own profits, how should I take their behavior
into account when making my own profit-maximizing de-
cisions? One of the solutions, which economists use to
answer this question, is GT. The application of GT has
been an important devel opment in mi croeconomics [7].
One can imag ine two different outcomes . First, the firms
might get together and form a cartel, coordinating their
behavior as if they are a single monopoly. Second, they
might behave independently, each trying to maximize its
own profit while somehow taking account the effect of
what it does on what the other firms do. This paper deals
with generation reliability evaluation of these.
Market demand curve has negative gradient, and the
amount of demand decrease is explained by “price elas-
ticity of demand”. This index is small for short terms,
and big for long terms; because in longer terms, custom-
ers can better adjust their load relative to price [7]. De-
mand function is generally described as P = a – b·Q.
Therefore, price elasticity of demand is explained as:
d
d
dQ
EPb

1
(1)
Let’s suppose load forecasted by dispatching center is
an independent power from price that equals to Qn. There-
fore, dem a nd function can be obt ai n ed as:
n
ndd
QQ
Pa bQbQbQEE
   (2)
Also Total Revenue (TR) is obtained as:
2
TRP Qa Qb Q
 (3)
2.1. Cooperative Behavior: The Cartel
Suppose all the firms decide to cooperate in their mutual
benefit: They calculate their co sts as if they were a single
large firm, produce the quantity that would maximize
that firm’s profits, and divide the gains among them-
selves by some prearranged rule. This condition is like a
single monopoly market.
Such a cartel faces a fundamental problem; it must
somehow keep the high price it charges from attracting
additional firms into the market. The cartel may try to
deter entry with the threat that, if a new firm enters, the
agreement will break down, prices will plunge and the
new firm will be unable to recoup its investment. How
might the cartel alter the situatio n? One way would be to
increase the entrance cost high enough. Therefore here,
it's considered that there is no new firm that enters the
monopoly market.
Offer curve of a firm, is part of the marginal cost (MC)
curve that is more than minimum average variable cost
[7]. Also total offer curve of all firms is obtained from
horizontal sum of each firm’s offer curve. This is a merit
order function.
In economics, if sale price in a market becomes less
than minimum average variable cost, the company will
stop production because it will not be able to cover not
only the fix cost but even the variable cost [9]. Due to the
changing efficiency and heat rate of power plants, mar-
ginal cost is le ss than average variable cost (AVC). There-
fore, in power plants, AVC replaces MC in economic stu-
dies [8].
In a monopoly market, the monopolist considers the
production level that maximizes his profit. It has been
proven that the monopolist considers the level of produc-
tion in which marginal cost of each firm (and total mar-
ginal cost of all firms) equals to the marginal revenue
(MR) of the monopolist [7]:
12
M
CMC MCMR
  (4)
where:


2
2n
dd
TRP QQQ
MRab Q
QQ E

E
 
 (5)
A typical total offer and marginal revenue curves are
shown in Figure 2.
2.2. Non-Cooperative Behavior: Nash Equilibrium
Another outcome in oligopoly power market is to assume
Copyright © 2012 SciRes. SGRE
Generation Reliability Evaluation in Deregulated Power Systems Using Game Theory and Neural Networks 91
MR curve
Total offer curve
P (mills/kWh)
Q (kW)
Figure 2. Typical total offer and MR curves.
that the oligopoly firms make no attempt to work to-
gether. Perhaps they believe that agreements are not
worth making because they are too hard to enforce, or
that there are too many firms for any agreement to be
reached. In such a situation, each firm tries to maximize
its profit, independently. If each firm acts independently,
the result is a Nash Equilibrium (NE).
Let i
s
be the strategy of player i, and i
s
be the
vector of strategies of all other players. Lets
,
ii i
uss
be the payoff to player I, then NE is a vector
,
ii
s
s
such that [6]:

,,;
ii iii ii
ussusss i

,
 (6)
That is, a NE is an outcome in which each player
chooses his strategy to maximize his payoff, given the
equilibrium strategies of all other players.
There are n firms, each selling an identical product on
a market with inverse demand function P(Y); where,
1
n
i
j
Y
y is aggregate output. Firm i has cost function

ii
Cy. Firms choose output, and choices are made si-
multaneously.
The problem for the firm i is:

1
max
i
n
j
iii
yj
PyyCy



(7)
which can be rewritten as:

max
i
n
ijii
yji
Pyy yCy




i
(8)
Since decisions are made simultaneously, firm i’s choice
cannot affect the choices of other firms. Thus, firm i per-
ceives correctly that 0;
ij
yy ji . Thus, the best
choice for the firm i is obtained using:

n
iiii
ji
PyyyP Cy





Equation (9) can be interpreted as a best response
function or reaction function for the firm i. It specifies
the best choice for the firm i in response to (or in reaction
to) the choices by other firms. This terminology is some-
what misleading since the firm i does not respond to the
actions of other firms in a sequential sense (since all
firms act simultaneously); rather firm i responds to what
it expects other firms to do.
How are those expectations formed? The firm i expects
all the other firms to play the strategy (output choice) that
is a best response to its choice.
Therefore, Cournot-Nash equilibrium

y
is charac-
terized by:

n
iiii
ji
PyyyP Cy





i
 
(10)
This is just a MRi = MCi condition. In other words, in
Cournot model, each firm, supposing that other firms
continue their present productions, acts as a monopolist.
Therefore, demand curve for the firm i is obtained as:
111
;
iiin
ii
PabQQ QQbQ
abQin


i


 
 (11)
where:
111iii
aabQ QQQ

n


(12)
Using (11), TR and MR of firm i are obtained as (13)
and (14), respectively:
iii ii
TRPQab QQ
 i
(13)



2
2
ii i
i
ii
ii
aQ bQ
TR
i
M
Ra
QQ

bQ

  (14)
Therefore, in an oligopoly power market with non-co-
operative behavior, generated power of each power plant
is obtained using the follo wing solution simultaneously:
min max
;1
Subject to:
;1
ii
jij
MR MCin
PGPG PGjm
 
(15)
3. Algorithm of Generation Reliability
Assessment Using GT and NN
Generation reliability of a power system depends on many
parameters, especially on reserve margin, which is defined
as [9]:
Installed CapacityPeak Demand
%1
Peak Demand
RM
00
(16)
Now let’s evaluate generation reliability for the men-
tioned two outcomes: cooperative and non-cooperative.
i
(9)
Copyright © 2012 SciRes. SGRE
Generation Reliability Evaluation in Deregulated Power Systems Using Game Theory and Neural Networks
92
3.1. Generation Reliability Assessment in
Cooperative Condition (Monopoly Market)
As explained before, in cooperative outcome, the market
acts as a monopoly market. The algorithm of generation
reliability assessment in monopoly power pool market
using Monte Carlo simulation is as follows:
1) Calculate total offer curve of the power plants.
2) Select a random day and its load (Qn), and calculate
MR curve using (5).
3) The power plants, selected for generation in the se-
lected day, are determined from the intersection of the
power plants’ total offer curve and MR cu rve with regard
to the reserve margin.
4) For each power plant selected in the previous step, a
random number between 0 - 1 is generated. If the gener-
ated number is more than the power plant’s Forced Out-
age Rate (FOR), the power plant is considered as avail-
able in the mentioned iteration; otherwise, it encounters
forced outage and thus can not generate power. This
process is performed for all power plants using an inde-
pendent random number generated for each plant. Finally,
sum of the available power plants’ generation capacities
is calculated. If the sum becomes less than the intersec-
tion of the power plants’ total offer curve and demand
exponent curve, we will have interrup tion in th e iteration,
and therefore, LOLE will increase by as much as one unit;
otherwise, we will go to the next iteration. The algorithm
of available generated power and LOLE calculation for
each iteration in MCS is shown in Figure 3.
5) The Steps 2 to 4 are repeated for calculation of the
final LOLE.
3.2. Generation Reliability Assessment in
Non-Cooperative Condition (Cournot-Nash
Equilibrium)
The algorithm of generation reliability assessment in oli-
gopoly power pool market for non-cooperation outcome
using MCS is as follows:
1) Select a random day and its load (Qn), and calculate
demand curve cross of basis and gradient using (2).
2) Calculate the power plants’ generated powers in the
selected day using (11)-(15). Also, the amount of total
demand is obtained using sum of plants’ generated pow-
ers regard l es s of rese r v e margi n .
3) For each power plant selected in the previous step,
with regard to the reserve margin, a random number be-
tween 0 - 1 is generated. If the generated number is more
than the power plant’s Forced Outage Rate (FOR), the
power plant is considered as available in the mentioned
iteration; otherwise, it encounters forced outage and thus
can not generate power. This process is performed for all
power plants using an independent random number gen-
erated for each plant. Finally, sum of the available power
Generate a random number
between [0-1] (U
i
)
Select the first generator (i = 1);
Available Generated Power (AGP) = 0;
LOLE = 0
AG P = AGP + P G
i
U
i
>= FOR
i
i = NG
i = i + 1
Total AGP calculation
AGP < Load
LOLE = LOLE + 1 LOLE does not change
N
N
Y
N
Y
Figure 3. The algorithm of available generated power and
LOLE calculation for each iteration using MCS.
plants’ generation capacities is calculated. If the sum
becomes less than the intersection of the power plants’
total offer curve and demand exponent curve, we will
have interruption in the iteration, and therefore, LOLE
will increase by as much as one unit; otherwise, we will
go to the next iteration.
4) Steps 1-3 are repeated for calculation of the final
LOLE.
3.3. Generation Reliability Evaluation Using NN
Now, to create a unique structure, a four-layer Perceptron
NN is used for reliability evaluation. Th e number of neu-
rons in each layer is 20, 15, 10 and 1, respectively (Fig-
ure 4). All neurons in the first, third and last layers have
POSLIN transfer function, and the second layer has
TANSIG transfer function. Three inputs of the NN in-
clude:
C: A number that shows the kind of outcome (1 for
cooperative, and 2 for non-cooperative outcomes),
Ed: Price elasticity of demand,
RM: Reserve margin.
Also, the NN’s output is LOLE index.
Copyright © 2012 SciRes. SGRE
Generation Reliability Evaluation in Deregulated Power Systems Using Game Theory and Neural Networks 93
Figure 4. Proposed NN for generation reliability assessment.
Parts of the MCS results, obtained from the mentioned
algorithm, are used for NN training.
4. Numerical Studies
IEEE-Reliability Test System (IEEE-RTS) is used for
case studies. Data for IEEE-RTS can be found in [10]. In
all case studies, the following assumptions are applied:
1) All case studies are simulated for the second half of
the year based on the daily peak load of the mentioned
test system.
2) All simulations are done with 5000 iterations.
3) Each study is simulated for four different reserve
margins (0%, 4.8%, 9% and 13%).
4) All scenarios are simulated for two price elasticity
of demands (0.001 and 0.01).
5) In Cournot-Nash outcome, it is assumed that each
power plant belongs to an independent firm (n = m).
6) NN is trained with TRAINLM method in MATLAB
software with 150 epochs. In this research, the NN reached
0.1 Mean Square Error (MSE) after training.
In the first study, price elasticity of demand equals
0.001. Based on this assumption and using MCS algo-
rithm and the proposed NN, LOLE values are obtained
versus different reserve margins as shown in Figures 5
and 6, respectively.
In the second study, price elasticity of demand equals
0.01. Based on this assumption and using MCS algorithm
and the proposed NN, LOLE values are obtained versus
different reserve margins as shown in Figures 7 and 8,
respectively.
As shown, in both case studies, LOLE values in the
NN method are very similar to those of the MCS method.
Evidently, the NN’s specifications depend on the power
system’s characteristics, and the proposed NN is valid for
the mentioned power system. Therefore, NN’s specifica-
tions may be changed in another power system based on
the power system’s parameters.
0
50
100
150
OUTCOM E
LOLE
[Days / S econd half of ye ar]
RM=0% 32.76 138.62
RM=4.8%25.36114.17
RM=9% 20.08 97.85
RM=13% 15.1759.15
Monopoly Cournot-Nash
Figure 5. LOLE values for the first study using MCS.
0
50
100
150
OUTCOME
LOLE
[ Day s / Sec ond half of year
]
RM=0% 32.7 138.84
RM=4.8% 25.18 114.15
RM=9% 20.08 97.8
RM=13% 15.36 59.16
Monopoly Cournot-Nash
Figure 6. LOLE values for the first study using NN.
0
50
100
150
OUTCOME
LOLE
[Days / Sec o nd half of y ear ]
RM=0% 27.85 137.89
RM=4.8% 24.39 104.77
RM=9% 15.71 79.41
RM=13% 14.74 70.55
Monopoly Cournot-Nash
Figure 7. LOLE values for the second study using M CS.
0
50
100
150
OUTCOME
LOLE
[Days / Second half of year
]
RM=0%27.8 138.12
RM=4.8% 24.38 104.76
RM=9% 15.94 79.42
RM=13%14.78 70.55
Monopoly Cournot-Nash
Figure 8. LOLE values for the second study using NN.
Copyright © 2012 SciRes. SGRE
Generation Reliability Evaluation in Deregulated Power Systems Using Game Theory and Neural Networks
Copyright © 2012 SciRes. SGRE
94
In both case studies, if reserve margin increases, LOLE
will decrease and reliability will improve.
In monopoly market, if price elasticity increases, MR
curve takes less gradient. As a result, intersection of the
power plants’ total offer curve and MR curve occurs at
less demand. This leads to the operation of fewer power
plants. Therefore, in all case studies in monopoly market,
if price elasticity increases, LOLE will decrease.
In Cournot-Nash equilibrium, if price elasticity varies,
the generated power of every power plant varies, too.
Therefore, LOLE will differ based on the share of every
plant’s generated power and FOR.
LOLE values in Cournot-Nash outcome are very big-
ger than those of the monopoly outcome. Because in mo-
nopoly market, only the plants, which are selected by in-
tersection of the total offer and MR curves, are in service
(considering RM), while in Cournot-Nash outcome, all of
the players participate in the market, and the load feeds
based on the plants’ optimum generated power. Therefore,
in monopoly market, at every time period, only a few
plants are in service, while in Cournot-Nash ou tcome, all
of the power plants are in service based on their optimum
generation. Since the number of in-service plants in
Cournot-Nash outcome is very bigger than in monopoly
market, generation reliability in monopoly market is bet-
ter than in Cournot-Nash outcome.
It is worth noting that, since available capacity of hy-
dro plants in IEEE-RTS are different in the first and the
second halves of the year, therefore, in the present work,
simulations were done for the second half of the year.
Evidently, the proposed method can be utilized for every
simulation time.
5. Conclusions
This research deals with generation reliability assessment
in power pool market using GT. Due to the stochastic
behavior of market and generators’ FOR, MCS was used
for simulations. Also, for creation of a unique structure
for reliability assessment, a NN was used, which its out-
puts were very similar to the MCS results.
Based on the players’ cooperation conditions, two out-
comes (Monopoly and Cournot-Nash equilibrium) were
considered. LOLE was used as generation reliability in-
dex, and it was shown that generation reliability in mo-
nopoly market is better than Cournot-Nash model.
Also, in monopoly market, if price elasticity increases,
LOLE will improve. On the other hand , in Cournot-Nash
model, if the price elasticity of demand varies, the power
plants’ generated powers will vary too, and that is why
LOLE changes.
REFERENCES
[1] R. Billinton and R. Allan, “Relia bility Evaluation of Power
Systems,” 2nd Edition, Plenum Press, New York, 1996,
pp. 1, 10, 372.
[2] L. Salvaderi, “Electric Sector Restructuring in Italy,” IEEE
Power Engineering Review, Vol. 20, No. 4, 2000, pp. 12-
16. doi:10.1109/39.833009
[3] H. B. Puttgen, D. R. Volzka and M. I. Olken, “Restructur-
ing and Reregulation of The US Electric Utility Industry,”
IEEE Power Engineering Review, Vol. 21, No. 2, 2001,
pp. 8-10. doi:10.1109/39.896811
[4] R. Azami, A. H. Abbasi, J. Shakeri and A. F. Fard, “Impact
of EDRP on Composite Reliability of Restructured Power
Systems,” IEEE PowerTech, Bucharest, 28 June-2 July
2009, pp. 1-8. doi:10.1109/PTC.2009.5281812
[5] P. Wang, Y. Ding and L. Goel, “Reliability Assessment
of Restructured Power Systems Using Optimal Load Shed-
ding Technique,” Generation, Transmission & Distribu-
tion, Vol. 3, No. 7, 2009, pp. 628-640.
[6] K. Leyton-Bro wn and Y. Shoham, “Essentials of Game The-
ory,” Morgan & Claypool Publishers, USA, p. 11.
[7] R. S. Pindyck and D. L. Rubinfeld, “Microeconomics,”
5th Edition, Prentice Hall, Saddle River, 2001.
[8] S. Borenstein, “Understanding Competitive Pricing and
Market Power in Wholesale Electricity Market,” Energy
Institute, University of California, Berkeley, 1999.
[9] International Energy Agency (IEA), “Security of Supply
in Electricity Markets—Evidence and Policy Issues,” IEA,
Paris, 2002, p. 16.
[10] IEEE Reliability Test System Task Force of Application
of Probability Methods Subcommittee, “IEEE Reliability
Test System,” IE EE Transactio ns on Powe r Appa ratus an d
Systems, Vol. 98, No. 6, 1979, pp. 2047-2054.
Generation Reliability Evaluation in Deregulated Power Systems Using Game Theory and Neural Networks 95
Symbol List
MC: Marginal cost (mills/kWh) (1 mills = 0.001 $)
TR: Total revenue (mills/h)
MR: Marginal revenue (mills/kWh)
AGP: Avail able gener ated powe r
Ui: A random number between [0 - 1]
NG: Number of power plants
Q: Quantity of power (kW)
P: Electrical energy price (mills/kWh)
RM: Reserve margin (%)
Ed: Price elasticity of demand (kW2h/mills)
C: Type of outcome
Qn: Forecasted load (kW)
LOLE: Loss of load expectation (days/second half of
the year)
FOR: Forced outage rate of power plants
a: Demand exponent curve cross of basis (mills/kWh)
b: Demand exponent curve gradient (mills/kW2h)
m: Number of power plants in the pool market
n: Number of independent firms in the pool market
PG: Generated power (kW)
PGmin: Minimu m limit of generator (kW)
PGmax: Maximum limit of generator (kW)
Copyright © 2012 SciRes. SGRE